Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77020.71829020776,
-      "required_pages_per_second_upper_estimate": 0.005328472004004809
+      "required_token_positions_per_second": 76910.27341947646,
+      "required_pages_per_second_upper_estimate": 0.005320831171580212
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9627.58978627597,
-      "required_pages_per_second_upper_estimate": 0.004030870796577202
+      "required_token_positions_per_second": 9613.784177434558,
+      "required_pages_per_second_upper_estimate": 0.004025090676449198
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19255.17957255194,
-      "required_pages_per_second_upper_estimate": 0.004010346111729625
+      "required_token_positions_per_second": 19227.568354869116,
+      "required_pages_per_second_upper_estimate": 0.00400459542324307
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28882.76935882791,
-      "required_pages_per_second_upper_estimate": 0.0017314779216264782
+      "required_token_positions_per_second": 28841.352532303674,
+      "required_pages_per_second_upper_estimate": 0.0017289950461161826
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28882.76935882791,
-      "required_pages_per_second_upper_estimate": 0.00224550827577266
+      "required_token_positions_per_second": 28841.352532303674,
+      "required_pages_per_second_upper_estimate": 0.0022422882996837677
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28882.76935882791,
-      "required_pages_per_second_upper_estimate": 0.3704771534334848
+      "required_token_positions_per_second": 28841.352532303674,
+      "required_pages_per_second_upper_estimate": 0.3699459028527555
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10242736661734853,
-    "producer_p95_seconds": 0.2923941859990009,
-    "gpu_update_seconds": 0.9572489277780938,
-    "producer_batches_per_second": 9.763015813301463,
-    "gpu_batches_per_second": 1.0446603500733476,
-    "producer_headroom": 9.3456364191634,
+    "producer_batch_seconds": 0.1052034928987382,
+    "producer_p95_seconds": 0.2854105980004533,
+    "gpu_update_seconds": 0.958623558622396,
+    "producer_batches_per_second": 9.505387819799221,
+    "gpu_batches_per_second": 1.0431623456417707,
+    "producer_headroom": 9.112088697901909,
     "sustainable": true,
-    "buffer_seconds": 60.306682450019906,
+    "buffer_seconds": 60.393284193210945,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323883.7922203457,
-    "upload_bytes_per_second": 6166064.001694573,
+    "generation_bytes_per_second": 323608.27983251715,
+    "upload_bytes_per_second": 6200102.377780693,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.49792218168047,
+    "estimated_full_shards_per_hour": 69.43880363685261,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.2500933689989324,
-      "bytes": 4989357,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.2095713759990758,
       "bytes": 7770394,
@@ -562,6 +556,12 @@
       "bytes": 2519802,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.0123512749996735,
+      "bytes": 5925157,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -271,3 +271,4 @@
 {"step": 512100, "loss": 0.6296825408935547, "lr": 0.0003, "grad_norm": 0.19045792368815628, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74954321694, "tokens_per_second": 151781.22230199672, "data_wait_s": 1.9504001102177426e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 512200, "loss": 0.6029542684555054, "lr": 0.0003, "grad_norm": 0.1968340038268829, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74968801634, "tokens_per_second": 150563.22624656284, "data_wait_s": 1.691799843683839e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 512300, "loss": 0.5672863721847534, "lr": 0.0003, "grad_norm": 0.17085796645715126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74983471338, "tokens_per_second": 152572.5015712951, "data_wait_s": 1.5751000319141895e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 512400, "loss": 0.5869703888893127, "lr": 0.0003, "grad_norm": 0.15479998636327552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74998130872, "tokens_per_second": 152838.09116581045, "data_wait_s": 9.548999514663592e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7be1b4f9dc84dff909f2333f3f3e572c6b04f772475ea5d574176d81a47ccf09
-size 582294
+oid sha256:f1a3f4b71f0a63c87077b2150ccf5674b16ca670387dd42d04de76cc21e470c1
+size 584412